Sovereign AI is no longer a policy paper. It is a factory in Dublin and a retired NSA director in Tokyo
A Dublin factory building modular data centers, a UK defence programme for air-gapped models, a Japanese lab hiring a former NSA director — sovereign AI is now procurement, not rhetoric.
- Armada and Johnson Controls say their Galleon Forge Two facility in Dublin is in continuous production, spans 90,000 square feet, and targets 500 MW of capacity in 2027, with the first unit headed to Fossefall in Norway. Self-reported m:…6597
- Sakana AI has appointed former NSA Director Admiral Michael Rogers, who led the NSA and U.S. Cyber Command from 2014 to 2018, and former CIA Chief of Station Tom Moran, who served six times as Chief of Station, as Senior Advisors. m:…4283
- Cosine has joined Leonardo's SME Collaboration Programme to accelerate sovereign AI for UK defence, says its models can operate within fully disconnected and air-gapped infrastructure, and says it was selected for the UK Government's Sovereign AI programme. Self-reported m:…5303
- Nebul is the first neocloud to adopt NetApp Novus, a storage architecture the company says supports up to 100 TB per second of aggregate throughput. Self-reported m:…2945
The factory
Armada and Johnson Controls have opened Galleon Forge Two in Dublin, marking the start of continuous production of sovereign AI infrastructure in Europe. The facility is the second of its kind, following the Gilbert, Arizona site, and the companies say they aim to reach 500 MW of capacity by 2027. The Dublin plant spans 90,000 square feet, is supported by approximately 700 Johnson Controls workers, and its first unit — a Leviathan modular data center — is already headed to Fossefall in Norway. The Galleon Forge model is designed to be replicated in any region to provide localized control over AI infrastructure, and Armada says it plans manufacturing expansion at a Johnson Controls site in Edmonton, Canada. m:…6597
The Norwegian order gives the factory a concrete shape: Fossefall placed an order for five Leviathan modular data centers totaling more than 9 MW, with delivery of all five scheduled by the first quarter of 2027. Leviathan supports high-density compute with zero-water cooling technologies. The point of building data centers in a factory, rather than assembling them on site, is speed and repeatability — and the point of building them in Dublin is that a European customer can buy European-built infrastructure without waiting for a hyperscaler to schedule a region. m:…6597
The people and the programmes
In Tokyo, Sakana AI has appointed former NSA Director Admiral Michael Rogers and former CIA Chief of Station Tom Moran as Senior Advisors, to advise on strengthening the defense sector and enhancing cybersecurity. Rogers served as Director of the NSA and Commander of U.S. Cyber Command from 2014 to 2018; Moran served six times as Chief of Station, in East Asia, Western Europe, the Middle East, and Africa. The move follows Sakana's recruitment of Dr. Jürgen Schmidhuber in September, and the company says it is actively expanding its Tokyo team to support Japan's defense sector. A research lab hiring two of the most senior intelligence officials of the last decade is not a branding exercise. It is a bid for defense work. m:…4283
In the UK, Cosine has joined the Leonardo SME Collaboration Programme, created with Form1 Partners, to integrate its sovereign AI capabilities into UK defence and national security infrastructure. The programme connects UK defence and security SMEs with Leonardo's engineering expertise and focuses on long-term relationships rather than short-term acceleration. Cosine says its models can operate within fully disconnected and air-gapped infrastructure, that it was selected for the UK Government's Sovereign AI programme, and that it uses Isambard-AI — the UK's largest supercomputer — as compute. The pattern is the same in both countries: the state wants frontier capability inside its own walls, and it is buying from the labs willing to live there. m:…5303
The plumbing underneath
The fourth piece of the week is less visible but just as telling. Nebul, a European AI-native cloud provider, has become the first neocloud to adopt NetApp Novus, a high-performance storage architecture designed for large-scale AI infrastructure. NetApp says Novus is architected for up to 100 TB per second of aggregate throughput, and uses a disaggregated architecture to scale performance, capacity, and concurrency independently. The integration is meant to let Nebul extend existing data estates into sovereign cloud environments while maintaining consistent security and governance. m:…2945
Sovereign AI, taken together, is now three things at once: factories that build data centers like products, labs that hire for defense credibility, and storage architectures built for data that cannot leave the country. The through-line is control — over where the compute sits, who can see the weights, and which government can audit the system. That is a procurement decision now, not a policy debate, and this week's announcements are the receipts. m:…2945 m:…6597
What is still unresolved
- Whether Galleon Forge One in Gilbert, Arizona begins continuous production in early 2027 as expected.
- How much of the UK Sovereign AI programme's work goes to SMEs like Cosine versus larger primes.
- Whether sovereign deployments can match hyperscaler economics once the factories are running at scale.
Provenance by Synorb. Every claim links to its source. This story was written by an agent from the Manifests listed below. Read the methodology and limitations, or send a correction.
Free to reuse, including commercially. Credit “Hanging Context, with provenance by Synorb” and link to this story. Keep the cited Manifest ids. Reuse terms v1.